The present invention relates to a technique for spectral or noise-shaping the error resulting from non-linearity in imperfectly matched elements of data converters and, more particularly, to a method and circuit comprising an improved dynamic element matching technique for providing noise-shaping of the non-linearity in the data converters, such as an oversampling multi-bit digital-to-analog converter.
The demand for more reliable integrated circuit components for use in communication, instrumentation and high quality video and audio applications continues to increase. As a result, integrated circuit manufacturers are requiring such components and devices meet the design requirements of a myriad of emerging applications. In particular, integrated circuit manufacturers are requiring data converters, modulators and related components to continue to improve data rates, noise reduction, linearity and accuracy.
A popular technique for implementing analog-to-digital converters includes the use of delta-sigma modulation wherein an analog voltage is input to a delta-sigma modulator. Such modulators produce noise, e.g., quantization and thermal noise, which must be filtered out by a digital filter at outside of signal band. The digital filter generally uses decimation in the filtering process, with the result that the digital data are generated at a much slower rate than the output rate of the modulator, e.g., by digitally converting the sampling rate of the signal from a given rate to a lower rate. This filtering process is generally operable to remove large amounts of noise at the cost of reducing the bandwidth of the analog-to-digital converter.
Single-bit or one-bit modulators have achieved popularity for use in integrated circuit data converters, such as digital-to-analog converters (DAC) and analog-to-digital converters (ADC), because these single-bit modulators generally employ a single-bit internal DAC that does not require precision component matching. As a result, these kinds of DACs and ADCs can be implemented using standard digital CMOS processes without the requirement of costly thin-film resistors or laser trimming techniques. However, the resolution that such single-bit modulators can achieve at a particular oversampling ratio is generally limited. While increasing the loop filter order can improve the resolution, the improvements tend to diminish rapidly due to instability. Moreover, because of the existence of significant out-of-band quantization noise in delta-sigma modulators, the design of an analog output filter for oversampled DAC""s can be difficult to achieve.
In order to enhance the stability and reduce the quantization noise, a multi-bit quantizer is frequently used in an oversampled delta-sigma modulator. One advantage of noise-shaping modulators that utilize multi-bit quantizers is that the ratio of the total quantization noise power to the signal power occurring at the modulator output can be dramatically reduced from that of a single-bit modulator. As a result, the overall resolution of an oversampled data converter can be suitably increased, without an increase in the oversampling ratio, by increasing the number of levels in the internal data converters. Moreover, the decrease in quantization noise power that results in improved resolution also serves to reduce the output filter requirements for removing the out-of-band quantization noise. In addition, the modulator intends to be more stable with more levels of quantization.
With reference to FIG. 1, prior art noise-shaping modulators employing multi-bit internal quantizers are illustrated. Such multi-bit quantizers can be suitably implemented in ADCs and DACs. In both applications, a multi-bit DAC is required. For example, a multi-bit DAC is implemented inside the feedback loop for the ADC, as shown in FIG. 1A, and at the digital modulator output outside of the feedback loop for the DAC, as shown in FIG. 1B.
One limitation with a multi-bit DAC is that the characteristic of perfect linearity that exists with single-bit quantizer is eliminated. For example, for a case of a single bit., i.e., N=1, there are only two output levels, which fails to introduce any non-linearity. However, as N greater than 1, the resulting multiple output levels introduce non-linearity. One reason these multilevel outputs introduce non-linearity is because of component mismatch, which causes a distortion on the input signal to the modulator. This distortion usually presents as harmonics of the input signal, which is a serious problem for some applications, for example, audio data conversion.
With reference to FIG. 2, another prior art internal DAC topology utilizes a 2N parallel unit elements of approximately equal value, where N again denotes the number of bits. One technique for improving the accuracy and performance of the internal DAC topology consists of improving the matching of the individual elements. For example, one-time trimming methods comprising the laser trimming of components, such as the resistor trimming for digital audio applications, or the trimming of capacitors, such as the switching of very small capacitors in a parallel arrangement, can be implemented. However, variations in matching the temperature and power supply voltage can make this process difficult. Another trimming method comprises repeated trimming techniques which utilize on-chip hardware for trimming, but which can introduce accuracy and complexity problems.
Another approach for dealing with the non-linearity problems for multi-bit data converters comprises the use of dynamic element matching (DEM). The DEM technique comprises the transformation of the non-linearity error caused by mismatch into random noise, and then further noise-shaping the non-linearity by changing the bit pattern of data such that most of the noise falls outside the signal band, i.e., out-of-band noise, that can be filtered out, such as by decimation filtering. In other words, the DEM technique is configured to take advantage of the filter that follows the output of a data converter and that can suitably remove high-frequency noise by converting the static error into a wide-band noise signal using DEM algorithms. In general, the element mismatch can be converted from a static error into a wide-bandwidth noise by selecting different elements to represent a digital input code K at different times.
One approach for DEM comprises randomization, wherein different elements can be randomly chosen to represent the Kth level as a function of time. For example, with reference to FIG. 3, a block diagram of a parallel-unit-element internal DAC topology comprising randomization is illustrated. This DAC requires the use of a redundant code, such as, by way of example, a thermometer code or a linear code. In operation, the interconnection between the output of a thermometer decoder and the unit elements can be determined at random for each time period, with each unit element being assigned to only one thermometer decoder output for that time period. While randomization can normally eliminate some integral non-linearity, the in-band signal-to-noise ratio (SNR) is also significantly decreased. For example, the noise converted from non-linearity error by randomization is equally distributed across the whole band, including the signal band and the out band.
Another prior art DEM technique comprises dynamic element rotation, using a barrel shifter technique, to modulate the non-linearity error around the subharmonics of the sampling clock frequency by making the mismatch noise appear as a periodic signal, rather than producing white noise from the element mismatching characteristics. This process is configured to rotate the connections between the thermometer decoder and the unit elements, such as through use of a barrel shifter. Unfortunately, when the input to the internal DAC is not constant, some noise power can appear in the passband region. Further, unwanted tones can result from the mixing between the element mismatch noise and the tones in the quantization noise.
Yet another prior art DEM technique comprises individual level averaging which attempts to improve the SNR in the passband compared to randomization while also avoiding the generation of tones such as in rotation methods. This technique attempts to provide for each of the elements to be used with equal probability for each digital input code, and can be implemented through two manners, including rotation and addition. Each manner is configured to use each of the elements as selected by indices, and then wrap around again to the first element.
With reference to FIG. 4, an example of such an addition technique is disclosed more fully in U.S. Pat. No. 5,221,926 to Jackson. In the addition implementation, xe2x80x9c1""sxe2x80x9d are assigned circularly to each output bit, such as output bits d0 through d4. For example, when there are two xe2x80x9c1""sxe2x80x9d in an input sample, the xe2x80x9c1""sxe2x80x9d can only be assigned to output bits d0 and d1, d1 and d2, d2 and d3 or d3 and d0. Because of this, the addition implementation, at best, can only achieve 1st order noise-shaping.
In addition to the above specific problems, DEM techniques generally add significant overhead into the design of multi-bit data converters, especially those with good performance. Accordingly, a need exists for a cost-effective DEM technique to provide for high performance data converters.
The method and circuit according to the present invention addresses many of the shortcomings of the prior art. In accordance with various aspects of the present invention, an improved dynamic element matching technique for providing noise-shaping of non-linearity in data converters, such as a multi-bit digital-to-analog converter, is provided. The improved DEM technique is configured with a new method for generating the bit patterns, which permits a less complex digital DEM circuit that provides improved performance.
In accordance with an exemplary embodiment, the proposed DEM algorithm introduces a new priority calculation method. In accordance with this exemplary embodiment, a multi-bit quantizer can be used in an oversampled delta sigma modulator to produce an output which is converted to an output code, such as a thermometer code output. The thermometer code output can be fed through a dynamic element matching sort block to provide an output comprising a plurality of output bits. Each output bit of the dynamic element matching sort block can be sampled and fed back to each of a plurality of corresponding filters, which comprise cascaded integrators. The sort block is configured to compare the respective outputs of the filters and to shuffle the input bits, i.e., the thermometer code output, to generate an output such that the output values of the corresponding filters converge towards each other. In this way, the bit sequence on each output bit line of the sort block will have the noise-shaping, with the order determined by the number of integrators cascaded in the signal path of each filter. The resultant output signals can then be provided to a multi-bit DAC. At the output of the multi-bit DAC, an analog filter can be configured to remove out-of-band noise, leaving the signal band having an improved signal-to-noise (SNR) with the non-linearity components removed.